Week 7 #
Topic: The Canonical Correlation Analysis Family in Self-Supervised Learning
Keynote Speaker: Xunyi Jiang, Langtian Ma
Time: Aug 3, 19:30 - 21:00 pm
Venue: Lecture Hall 3, 302 (SUSTech)
Online Link: TencentMeeting
Compendium #
I. Traditional CCA
- Generalized CCA framework
- Traditional Nonlinear CCA
- A compressed representatoin approach for CCA
- Kernel CCA
II. Deep CCA and its variates
- Deep canonical correlation analysis
- Deep canonically correlated autoencoders
III. CCA in Self-supervised Learning
There are 4 major categories of self-supervise methods, including information maximization, clustering, distillation techniques, and contrastive method. In this week, I will introduce 3 methods(W-MSE/Barlow Twins/VICReg) lying in information maximization, and 2 methods(SeLa/SwAV) in clustering.
The story line is as following:
- Overview of SSL methods
- Information maximization methods: W-MSE/Barlow Twins/VICReg
- Clustering methods: SeLa/SwAV
- Summary of these methods
Material #
Slides 1 and 2 for Canonical Correlation Analysis Family from Xunyi Jiang and Langtian Ma.
References #
Breiman, L et al, Estimating Optimal Transformations for Multiple Regression and Correlation
Painsky A et al, Nonlinear Canonical Correlation Analysis:A Compressed Representation Approach
D. R. Hardoon et al, Canonical correlation analysis: An overview with application to learning methods
Andrew, G et al Deep canonical correlation analysis
Wang, W et al On deep multi-view representation learning
A. Ermolov, A. Siarohin, E. Sangineto, and N. Sebe, Whitening for Self-Supervised Representation Learning
J. Zbontar, L. Jing, I. Misra, Y. Lecun, and S. Deny, Barlow Twins: Self-Supervised Learning via Redundancy Reduction
A. Bardes, J. Ponce, and Y. Lecun, VICREG: VARIANCE-INVARIANCE-COVARIANCE RE- GULARIZATION FOR SELF-SUPERVISED LEARNING
Y. Asano, C. Rupprecht, and A. Vedaldi, SELF-LABELLING VIA SIMULTANEOUS CLUSTERING AND REPRESENTATION LEARNING
M. Caron, I. Misra, J. Mairal, P. Goyal, P. Bojanowski, and A. Joulin, Unsupervised Learning of Visual Features by Contrasting Cluster Assignments